DDraceNetwork - off-topic

ive gotten myself hooked on terraria again i s2g the ost is so goated

Voxel

ive gotten myself hooked on terraria again i s2g the ost is so goated

@keb

cyberFighter

@keb

terraira awesome

12:57

with like 6 people

keb

terraira awesome

would you rather play master mode or no wiki on

cyberFighter

would you rather play master mode or no wiki on

mastamod

14:33

they should add the no wiki gamemode

14:33

and its normal mode without a wiki

14:33

finalday

14:33

like it just

14:33

closes all messaging apps

14:33

and browser

i messed up a pll algorithm and it somehow worked anyways

15:32

wtf

pilonpl

i messed up a pll algorithm and it somehow worked anyways

15:44

from where do you learn em?

https://jperm.net/algs/2lookpll

15:45

nice website

i was learning by felix's vids xD

this captcha is highly subjective, but i think all images qualify to some people

17:25

i mean

17:25

except for the shirt on the top right

17:25

technically everything is edible in these photos

17:25

sure, some of these are more frowned upon or just illegal

17:25

but technically, you could tick any image here

middle pic 5000 calories

they say fruit as if there's any fruit here other than lemons

17:30

these captchas are getting wild

it looks like some new sort of training algorithm thing (edited)

17:31

ive noticed theyve changed in styles as well

they look AI generated

17:32

like

17:32

look at the bottle of oil

17:32

the text on the label pretty much confirms it

17:33

also, the captcha about "things you can eat" is absolutely, most certainly gonna be used to train AIs with morals

17:34

technically, nothing is stopping you from eating a bear

17:34

i mean (edited)

17:34

as long as it doesnt eat you first

ah yes

17:34

Fruit

my favourite Edible Object

17:35

I am a human

17:35

And i consume Such objects

would not recommend consuming objects

Depends on what you call an object

17:50

If you consider a slice of cake to be an object, I think it's fine

17:51

Or any food item for that matter

i wouldnt but idk the proper definition for an object

I'd be careful with something like defining a bowl of cereal as an object

17:51

Because that would imply the bowl is edible as well

17:51

Which uh

17:51

I don't think it is?

1. a material thing that can be seen and touched. "he was dragging a large object" synonyms: thing, article, item, piece, device, gadget, entity, body, thingamajig, thingamabob, thingummy, whatsit, whatchamacallit, what-d'you-call-it, thingy, doodah, doobry, gubbins, doodad, doohickey, doojigger, dingus, thingo 2. a person or thing to which a specified action or feeling is directed. "disease became the object of investigation" synonyms: target, butt, focus, recipient, victim 3. a noun or noun phrase governed by an active transitive verb or by a preposition. "in Gaelic the word order is verb, subject, object" 4. a data construct that provides a description of anything known to a computer (such as a processor or a piece of code) and defines its method of operation. "the interface treats most items, including cells, graphs, and buttons, as objects"

Well

17:55

"I ate an apple"

17:55

I think that's good enough

17:56

"I ate a bowl of cereal" on the other hand

17:56

That's a tricky statement

i still wouldn't call food objects

National Insurance Clerk

"I ate a bowl of cereal" on the other hand

Especially if I add "and it was crunchy"

bencie

i still wouldn't call food objects

Well then, I suppose in that case it really is not recommended to consume objects

17:58

But consume doesn't necessarily mean eat

17:58

You could consume media

17:58

Media is an abstract concept though, so probably not an object

different kinda consume but ye

18:00

all this started from some broken english lmao

Language is confusing when you look into it too closely

18:03

And yet even with all that ambiguity we can understand exactly what someone means

Wait is that a real thing?

18:21

I thought the AI was the so called artist in that context

18:21

And even then

18:21

It owes its existence to all the artists whose art it was trained on

u actually have to be decent at photoshop in some cases but yeah theyre not rly artists

National Insurance Clerk

It owes its existence to all the artists whose art it was trained on

eeeeeh the meaning of "trained on" is only well defined for an ai, most human artists also take incredible amounts of influence from their fellow artists, wouldn't it make sense to say that artists were trained on prior art in a way too? (edited)

Huh

How would you differentiate human art from ai art? Can you come up with what exactly the human adds to the equation?

Idk

20:06

But I suppose the difference is the work you put into it which really makes it special

20:08

Be it forming an image out of your heart, or programming an AI that's capable of doing that (code could also be considered some sort of art)

Omg, i missed an ai discussion?

However, typing in a prompt.. I don't know

20:09

Unless you go into extreme depths in the prompt to the point where you're almost describing the image like a piece of literature

20:09

I don't know, it just doesn't feel the same

It doesn't feel as valuable, yet the end products could be very similar, so similar that you may not be able to tell them apart really soon

That's true

Ai ofc doesn't have feelings or emotions the same way humans do

If it's feelings and emotions that the human adds, couldn't we consider that it's the prompt that has these? (edited)

Just like AI chatbots may soon be nearly indistinguishable from humans

pilonpl

Ai ofc doesn't have feelings or emotions the same way humans do

not yet

Learath2

If it's feelings and emotions that the human adds, couldn't we consider that it's the prompt that has these? (edited)

Again, depends on how much you really express it in the prompt

20:11

"an oil painting of a forest" doesn't really scream emotion to me

People assume ai is stupid when it behaves differently that humans even tho the entire point of ai is that it doesn't behave like a human

I think it just behaves almost like a human, and so that's the sort of expectation people have

National Insurance Clerk

"an oil painting of a forest" doesn't really scream emotion to me

Well what if we now consider the selection process aswell? That also adds real information to the end result

Selection?

pilonpl

People assume ai is stupid when it behaves differently that humans even tho the entire point of ai is that it doesn't behave like a human

the entire point of ai is to be like humans

You generate thousands of images, you pick one to put out there, you are in some way restricting the output space of the ai, the same way a prompt does. You in a way by manipulating the prompt and picking from the results impart your "feelings", "emotions", "ideas" into the result.

at least neural networks

I find it quite fascinating, it's really hard to pinpoint why AI art is "different", yet most people feel quite strongly about it

Learath2

I find it quite fascinating, it's really hard to pinpoint why AI art is "different", yet most people feel quite strongly about it

its called moral

20:15

https://en.wikipedia.org/wiki/Morality

Morality

Morality (from Latin moralitas 'manner, character, proper behavior') is the differentiation of intentions, decisions and actions between those that are distinguished as proper (right) and those that are improper (wrong). Morality can be a body of standards or principles derived from a code of conduct from a particular philosophy, religion or cu...

pilonpl

People assume ai is stupid when it behaves differently that humans even tho the entire point of ai is that it doesn't behave like a human

I wouldn't say it's stupid as much as its just not human. ChatGPT for example is capable of some very impressive things, but nothing about it feels truly human, with things like not being able to respond to questions about preference and opinion

Ryozuki

its called moral

Expand on that. Do you think AI art is not different but people feel strongly about it because of moral qualms?

20:16

Or it is different due to the fact that the AI lacks morals?

if they didnt knew it was by a AI they would probs judge it equally

National Insurance Clerk

I wouldn't say it's stupid as much as its just not human. ChatGPT for example is capable of some very impressive things, but nothing about it feels truly human, with things like not being able to respond to questions about preference and opinion

ChatGPT is nothing but very very very fancy autocomplete

they want to feel good about themselves, its social behaviour, others dont like it, others say it discredits artists, i will have a bias then

Learath2

ChatGPT is nothing but very very very fancy autocomplete

That statement applies to pretty much all language based AIs, doesn't it?

this reminds me of genetics engineering

20:18

on a tech standpoint, its the most efficient

20:18

but morally wrong

20:18

because humans say so

Sort of hard to explain why it feels that way

20:19

It just does?

morality is that

20:19

its what you feel "is correct, good, not bad"

Like eating cats for example

20:19

That captcha, lol

Morality may also be specifically synonymous with "goodness" or "rightness".

National Insurance Clerk

That statement applies to pretty much all language based AIs, doesn't it?

Exactly, that is how they are designed

Was I supposed to mark just the images of cake?

20:20

Because cats are also technically edible, it's just that most of us would not do that

lol have u watched gattaca

That's also a puzzling question

20:21

Why eat a cow but not a cat

20:21

There's no objective explanation as to why

Well cats are not very suitable for mass production of meat

I guess so

20:22

We decided that cats are pets, so..

20:23

Still, someone else might say that eating cats is no different than eating any other animal

20:23

What would they answer on that captcha?

20:24

Man I just wanted to log in from a different browser

National Insurance Clerk

Still, someone else might say that eating cats is no different than eating any other animal

because it technically isn't, morality is not universal by a long shot

Morality is subjective

I'm sure early humans would have no problem domesticating and eating cats if they were suitable for domestication and if they were herbivores. If that had happened, today we might have had a problem eating cows but no problem eating cats

(edited)

Although there may be widely agreed upon things, there could always be that one person who would say otherwise

I like to define a time and space local average morality and judge people according to that

are u saying there are no universal morals

A while back people were pretty chill with burning random people alive because they're witches

20:28

Or something

20:28

Nowadays, not so acceptable

but back then we also has slavery

Learath2

ChatGPT is nothing but very very very fancy autocomplete

Neural networks can approximate any function so doesn't that mean that they're "turing complete" or something? The idea that it's autocompleting is just one "model" of thinking about it i guess?

louis

are u saying there are no universal morals

Not really, any minimal set of morals you can come up with, there is probably a local community somewhere that disagrees with one or two things in it

Or just one guy with a twisted mind

louis

are u saying there are no universal morals

no

20:30

morals are set by each person, and HEAVILY influenced by the society you live in

Morals help us be organized and efficient, I suppose

20:31

If everyone was fine killing each other we probably wouldn't have functional societies

u know people put up with killing for just robbery back in time

20:31

so morals are quite flexible

Ryozuki

morals are set by each person, and HEAVILY influenced by the society you live in

wb something like genocide or extinction or even in futuristic terms things like destroying entire planets for no good reason

Well

pilonpl

Neural networks can approximate any function so doesn't that mean that they're "turing complete" or something? The idea that it's autocompleting is just one "model" of thinking about it i guess?

Turing complete is a pretty tight definition, a lot of things that you wouldn't consider remotely useful are turing complete

the problem is humans have complex problems, and they often dont have a truly correct solution

20:32

so each other things their correct solution

20:32

thinks

20:33

this reminds me of the trolley problem

20:33

xddd

Those committing genocide most certainly don't think it's wrong

louis

wb something like genocide or extinction or even in futuristic terms things like destroying entire planets for no good reason

There was a time when a sizeable enough subset of people thought genocide was fine that it happened. I'm sure you can find a tribe or two that would be completely okay with the genocide of their opposing tribe even today

I hate moral dilemmas because I can't justify any answer lol

Yeah that's tough

It's very annoying because i like having solid explanations of stuff i belive

Even a moral dilemma as simple as "does this shirt make me look fat?" No matter what you say, you're either a liar or a piece of crap

20:35

Well that's not quite a moral dilemma

20:35

I just think it's kinda funny

Learath2

There was a time when a sizeable enough subset of people thought genocide was fine that it happened. I'm sure you can find a tribe or two that would be completely okay with the genocide of their opposing tribe even today

i read an argument saying the more generally educated a species or group is, the more likely they're to be morally "correct"

It is of the same nature of having two answers, both wrong

20:35

Both are right and wrong really

louis

i read an argument saying the more generally educated a species or group is, the more likely they're to be morally "correct"

Correct that is aligned with what the author of the argument thinks? Lol

louis

i read an argument saying the more generally educated a species or group is, the more likely they're to be morally "correct"

but according to what set of morals is the question, or what even educated means

If you believe in honestly, you're a piece of crap If you believe in saying only kind things, you're a liar

20:36

It's a stupid ass example

20:36

But it kinda works

National Insurance Clerk

Both are right and wrong really

There are many paradoxes like "this statement is false"

id make the argument that there is a high set of morals per species

20:37

that is generally correct

20:38

so there would be a high set of morals for humans as a whole

How could you know what that "high set of morals" is though?

pilonpl

There are many paradoxes like "this statement is false"

I don't think this is quite the same as a moral dilemma. Moral dilemmas have no right or wrong answers because morals are subjective, while in these paradoxes, there's no truth value to assign because the logic makes no sense

Say the more educated a person is the more likely they are to get closer to this "high set of morals". There were many a decorated scientist on the Manhatten Project, they didn't feel bad about vaporising entire cities.

National Insurance Clerk

I don't think this is quite the same as a moral dilemma. Moral dilemmas have no right or wrong answers because morals are subjective, while in these paradoxes, there's no truth value to assign because the logic makes no sense

Well yeah

Learath2

How could you know what that "high set of morals" is though?

define it

Does that mean this "high set of morals" doesn't include don't commit mass murder? Or does that mean that science education does not count? If so what flavour of education is needed?

like any set of rules that leads to maximum happiness

thinking someone has higher morals is a moral issue itself

or your own definition

louis

like any set of rules that leads to maximum happiness

We can't know the future though. Maybe murdering 60% of the population will lead to much more relative happiness on earth due to the extra resources?

happiness is a moral issue too

louis

like any set of rules that leads to maximum happiness

i hate any discussions where terms like "maximum happiness" are used lol. Arguments with these terms feel just so wrong

Learath2

We can't know the future though. Maybe murdering 60% of the population will lead to much more relative happiness on earth due to the extra resources?

then that might as well be morally correct

20:43

but i dont think that would lead to more relative happiness

20:43

its kind of an absurd example

pilonpl

i hate any discussions where terms like "maximum happiness" are used lol. Arguments with these terms feel just so wrong

then use "utilitarian standpoint"

louis

then that might as well be morally correct

but would you be okay with the "higher set of morals" not containing "no mass murder"? I honestly wouldn't. My set of morals (and pretty sure the local average around me) does not condone mass murder (edited)

its a better definition anyways

Also happiness is an emergent property, idk if it's possible to measure it in a way that makes sense

Learath2

Does that mean this "high set of morals" doesn't include don't commit mass murder? Or does that mean that science education does not count? If so what flavour of education is needed?

logic is often cold and doesn't really have a regard for morals, I think. That's the fear most AI overlord scary stories portray - the over exaggerated AI overlord realizes that humans do not contribute anything, and it is therefore at it's best interest to get rid of them. It thinks this is the most logical thing to do, and it doesn't care about it being morally wrong because who said it was morally wrong to commit genocide against the entire human race, if not us humans?

Learath2

but would you be okay with the "higher set of morals" not containing "no mass murder"? I honestly wouldn't. My set of morals (and pretty sure the local average around me) does not condone mass murder (edited)

well thats why i said its a bit of an absurb example

20:45

maybe replace it with antinatalism when population is high

20:45

then id be fine with that

For something like an AI to have morals is a whole separate point, it's honestly much much more complex. What would it even mean for a machine that just makes statistical calculations to have "morals"? A bias in the tensors?

Ai doesn't have the same motivations humans do. The goal of life is to survive. Ai doesn't have such goal

Maybe performing horrible experiments is morally wrong, but it lets all of humanity become more advanced, so does the "greater good" override personal morals?

National Insurance Clerk

Maybe performing horrible experiments is morally wrong, but it lets all of humanity become more advanced, so does the "greater good" override personal morals?

This is a less absurd but nice example

I don't even care about "morality". I just do whatever i want l o l

I gave the fictional AI as an example, because I think it's one of the more widespread ways to portray the issue of the greater good overriding morals, I didn't mean to treat it literally

Idk, I just find it very very hard to even think about a set of morals that applies to everyone at any time period

Because there is no such thing as a global, correct set of morals

Learath2

Idk, I just find it very very hard to even think about a set of morals that applies to everyone at any time period

caveman era: help kill mammoth

pilonpl

I don't even care about "morality". I just do whatever i want l o l

I am probably a horrible person

20:49

Glados agrees

caveman era: @louis didn't help kill mammoth, kill @louis he only eat our food no contribute

pilonpl

Glados agrees

You mean funni potato robot?

Is killing moral? eeeeeeeeeeeh not in my eyes, but if I apply my set of morals to the caveman era I'm missing some very crucial context

Well

20:51

It is technically at their best interest to kill those who don't help gather food

If you want to kill somebody, just do it. You probably would also want to make sure police doesn't find you

Not morally okay by today's standards, but I don't think they care

pilonpl

I don't even care about "morality". I just do whatever i want l o l

u dont

pilonpl

I don't even care about "morality". I just do whatever i want l o l

I very much doubt that you lack a set of morals that you subscribe to, your grey area might be larger but unless you are clinically insane it's very unlikely you completely lack morals

I mean

20:52

I follow some rules i guess

Would you rob someone because it'd make you feel better if you knew you would never be arrested for it?

Maybe

i dont tgink u would

1

depends

20:53

most moral compasses change on the distance

20:53

to the affected

i think you like the edgy aesthetic

we dont rly care abotu africa

ah yea the whole africa thing

20:54

its funny

Ryozuki

most moral compasses change on the distance

Mine was a very concrete example. You get out of your house, you find person, do you rob him?

the "would u like to donate to feed a starving kid in africa blah blah"

for me depends

20:54

if he has a good looking life

20:54

not poor

20:54

i would

20:54

if i dont get policed

Ryozuki

i would

Even though you have no need for his money?

The main issue with stealing is that you can get in a lot of unnecessary trouble

if i have 0 need no

20:55

but society always needs money

20:55

xd

I don't think it makes sense for me to steal anything lol

I was about to say that this is where the relativism comes in, if you are starving, ofc you would and you might even decide that it's the moral thing to rob them

20:57

If you really would just rob a random person when you don't have an immediate need, you are not the average person. It's because the average person doesn't do that that society can function

Ryozuki

most moral compasses change on the distance

Take the example of the trolley problem where the train is about to hit 5 people, but you can stop it by pushing a big guy off a bridge It's a lot harder to make the 1 < 5 choice when you have to be so close to the one you're killing

20:57

Even the lever example

20:58

You either let 5 people die and have no relation to the situation in a way

20:58

Or you let 5 people live and actively lead to the death of a person by redirecting the train to them

20:59

I like the fat guy on a bridge example better because it really highlights how close you are to the murder of that one person

Aaaaanyway, too much ethics discussion for one day, I'll go try to sleep, need to wake up early tomorrow for once

im raiding vs golbez in ff

20:59

while talking

I think the fat man just would decide if he wants to sacrifice himself?

The man is.. idk.. he fell asleep on his shift or something

21:00

Or even worse, he doesn't want to sacrifice himself, and you forcefully push him for the greater good

21:00

But how much greater is it?

Idk why but it feels different from the normal variation of the trolley problem

Because again, in that example you're REALLY actively killing the one person for the sake of 5

Irl i would probably just start recording with my phone

Even though flipping a lever still has the same "technically actively killing a person" deal, it's less prominent because you still dont have much contact with the person you actively kill

And do nothing

You're just flipping a lever, so you could think you caused his death, but didn't directly kill him

21:03

However, in the bridge example, there's a lot more contact

21:03

You are indeed actively killing, there's no way to phrase it nicely to yourself anymore

21:04

Then there's the extreme dilemma of whether it's morally right to save the life of a person trying to end their own life, but I feel like it's not something suitable to discuss here

It's very hard to tell exactly what's different because irl you would have a lot more context like your family and actual law

21:06

In the thought experiment many things are missing

21:06

And you are supposed to assume a lot of stuff

That's also true

21:07

But it's morally wrong to set up these experiments for real and kill someone

I just can't think of a sensible explanation as to why i would choose one option over the other

At least by our current standards

pilonpl

I just can't think of a sensible explanation as to why i would choose one option over the other

Yup that's the thing with moral dilemmas

21:12

You can't find the words to explain your choices

21:13

It obviously depends on something in you, like past experiences, beliefs, etc, but it's hard to point out exactly what it is

I also don't like morality because idk how it relates to the fact that fundamentally everything is made of atoms

21:15

And also free will

21:15

If you don't have free will

21:16

Is it really your fault you have done something immoral?

21:17

And free will doesn't really make much sense at all

21:17

But we also kind of assume everyone has free will

21:17

And we assume they're responsible for what they do

Who knows

21:18

If there's anything I take away from this, its

21:19

Life is too short to find out - commit felonies

The fact that anything exists at all is scary

21:20

I hope our universe is just one time event

21:20

And it's gonna die

21:20

And never come back

21:20

Because

21:20

If it's some sort of a loop

21:20

Or there are infinite amount of universes

21:20

Then that's pretty shit actually

21:20

Or is it?

21:21

Maybe it's just a subjective experience

21:21

And it doesn't matter like at all

You know what's even wilder than a universe?

21:23

Boltzmann brain.

Omg

It's a really freaky concept

Yeah

21:24

So basically the whole universe might be a lie

21:24

It might have been created like a second ago in this exact configuration

21:24

And you have no way of proving that's not the case

21:26

Also

21:27

We have consciousness

21:27

Amd we are basically a bunch of atoms

21:27

So like it's just information i guess

21:28

Isn't it like anything could have some sort of a structure?

pilonpl

We have consciousness

No no no

21:28

You have consciousness

21:28

But you cannot prove anyone else has it

21:28

Well, from my POV I can only confirm my own consciousness but not anyone else's

Wait, so you are an actual human? Or at least you think you are?

21:29

And you have thoughts

21:29

:0

After all, AIs don't need consciousness in order to have a conversation, and a lot of the actions your body performs, like heartbeats or metabolism aren't done with conscious effort

21:30

Seems like you wouldn't really need a consciousness to function

21:30

And maybe you do

21:30

After all we are much more complex than robots

21:30

But my point is

21:32

Everyone could very well just be a much smarter, biology-based Sophia and you'd have no way of confirming or denying it

21:32

I'm specifically referring to a robot with an actual body and senses rather than something like our good friend glorified autocorrect

It's quite weird

21:33

Because

21:33

You could technically simulate consciousness

21:33

On a computer

How do you know?

But not even on a computer

21:34

There could be some random system

21:34

Of random atoms relating to eachother

We still don't know if consciousness is a biproduct of complexity or something beyond

That kinda makes consciousness

National Insurance Clerk

We still don't know if consciousness is a biproduct of complexity or something beyond

Well, it's probably just complexity

If it really is a biproduct of complexity, then yes, a complex enough simulation of a brain could be conscious

pilonpl

Well, it's probably just complexity

Nothing else makes much sense

But how would we know if our very complex brain simulation is really conscious?

21:36

It could tell us that it's conscious all it wants

But also you could represent the same computations many different ways

And yet, we wouldn't know for sure just like you can only assume me and everyone else are conscious

pilonpl

But also you could represent the same computations many different ways

Imagine how scary it is to be a brain without a body

21:37

Wait why did it reply

21:37

Didn't mean to reply to that

But isn't it a matter of perspective? Neurons process information and this information processing creates consciousness so like there might be systems that we are not even considering might be doing soem sort of computation that might be different tha our consciousness bur its still a system of sorts

21:39

It's hard to explain

21:40

The universe itself might be "conscious" or perform some type of computation that we can't even imagine

And yet, you can't prove or disprove consciousness in any of them

If you don't believe in a sort of soul that is metaphysical, it is indeed correct to say that a complex enough simulation can simulate a human consciousness and all

We know quite a lot about our universe and nothing really indicates that there should be something like a soul

If you can perfectly simulate the complete physical state of a human brain, that only leaves something that could be metaphysical

pilonpl

We know quite a lot about our universe and nothing really indicates that there should be something like a soul

And nothing really disproves it, nor can it be disproven. It's completely unscientific to talk about an unfalsifiable thing like this

If you can perfectly simulate the complete physical state of a human brain it might be able to perform calculations and stuff, but it would still be a bit of a huge jump to say it's conscious

21:44

It's something we just can't look through

21:44

It's not something you can measure

So my measure is that if something can convince me that it's conscious, it's conscious enough for me

Well idk

21:46

Chatgpt could definitely be considered conscious

21:46

It's different that a human brain but still

21:46

It does have some understanding

pilonpl

Chatgpt could definitely be considered conscious

Chatgpt can easily be shown to not have an actual understanding of anything it parrots out from it's massive dataset

Of concepts

If a glorified autocomplete that's EXTREMELY close in communication skills to a human, and without this stupid "as an AI language model Im not actually conscious yada yada" line, would you be convinced?

metaphysical soul goes hard imo

Damn that'd be sad, but also a relief

Ai says what it was trained to say

metaphysical soul is better anyways that way we don't get a bunch of computer people

pilonpl

Ai says what it was trained to say

Chatgpt probably could pretend it's a human but that's just not open ais goal

National Insurance Clerk

If a glorified autocomplete that's EXTREMELY close in communication skills to a human, and without this stupid "as an AI language model Im not actually conscious yada yada" line, would you be convinced?

It would have to show capability of actual understanding. I doubt it can reach that through language models as I understand them, I think those models are missing some deeper layer

No consciousness to AI = no feeling pain = no hate towards humanity = I'm not obligated to scream but I have a mouth

louis

metaphysical soul is better anyways that way we don't get a bunch of computer people

Oh yeah uploading your mind to a computer is one wild discussion

But yes, if it could show capacity to adjust it's beliefs at runtime, prove to me that it understands things and can introspect. I'd consider that conscious enough

Fair

Because that's the only definition of consciousness I have for now btw. If some philosopher or scientist manages to come up with some better measure of it maybe I could change my beliefs

I don't think consciousness is something we could ever truly grasp or measure

Learath2

But yes, if it could show capacity to adjust it's beliefs at runtime, prove to me that it understands things and can introspect. I'd consider that conscious enough

Chatgpt can generate responses that linguistically make sense and it also perfectly understands what you want it to do. Isn't that already impressive? Language is very complex and i think you have to have some sort of understanding in order to use it

National Insurance Clerk

I don't think consciousness is something we could ever truly grasp or measure

but is that a useful definition? if nothing can be classified as conscious but yourself I think there isn't much utility to such a definition

pilonpl

Chatgpt can generate responses that linguistically make sense and it also perfectly understands what you want it to do. Isn't that already impressive? Language is very complex and i think you have to have some sort of understanding in order to use it

It can be easily shown that what it does isn't so much as understanding what I want it to do, but being very good at predicting what words should come after one another

Oh by the way, something unrelated to consciousness

21:54

Try asking Clyde or ChatGPT to write some text in upside down letters

21:54

It absolutely breaks them

It'll happily commit the freshman's dream mistake in math proofs, I can correct it, I can uncorrect it, it's incapable of making any deduction. It does not understand so much as to why it's mistake is a problem but that it's just not supposed to use that path in the word space anymore

They say some barely intelligeble stuff, and when you ask them to translate it, they say something completely different

werent u going to sleep

btw, I'm not saying it's not extremely impressive, it just doesn't reach the level of sentience most people think it has

Ryozuki

werent u going to sleep

I'm in bed, but can't sleep :/

For some reason I find it a bit eerie, when I tried it with ChatGPT it added some swear words in too, except it spelled them like it was drunk

21:58

I mean, this text is kinda ok

Learath2

It can be easily shown that what it does isn't so much as understanding what I want it to do, but being very good at predicting what words should come after one another

Doesn't predicting next words require you to understand the meaning of previous words? Isn't this the exact thing we understand by "understanding"? Like yeah, it does "just" predict words but doesn't predicting words imply that it has understanding?

But when ChatGPT was writing upside down, it was talking like a drunk angry dude

pilonpl

Doesn't predicting next words require you to understand the meaning of previous words? Isn't this the exact thing we understand by "understanding"? Like yeah, it does "just" predict words but doesn't predicting words imply that it has understanding?

It knows patterns, like your autocomplete does

22:01

This guy named CaryKH made a little AI that generated rap lyrics, and for example one thing he pointed out is how the AI picked up on the fact that the word 'motherf***ing' is typically followed by a noun

22:01

Idk why this is the first example that came to mind, I just find it kinda funny

pilonpl

Doesn't predicting next words require you to understand the meaning of previous words? Isn't this the exact thing we understand by "understanding"? Like yeah, it does "just" predict words but doesn't predicting words imply that it has understanding?

When I give you the sequence of words "I love eating" and ask you to complete the sentence. You don't just go to your bank of words and check which one more frequently appears after that sequence

I love eating.

22:02

Just that

22:02

No continuation whatsoever

When you say the word apple, you have a deep understanding about what an apple is, you know many properties about it, I can't convince you that an apple isn't a fruit but actually an animal

Learath2

When I give you the sequence of words "I love eating" and ask you to complete the sentence. You don't just go to your bank of words and check which one more frequently appears after that sequence

Righttt, but i think it's just because i am a human and i do actually have preferences

All you need to know is that an apple is something you typically eat, don't you?

I can assure you I can convince chatgpt of that, it doesn't understand what an apple is, so it'll happily adjust it's weights around to tell me it's an animal the next time I ask in that same chat window

Lol

22:03

Gaslighting ChatGPT

22:03

It's quite amusing

Learath2

When you say the word apple, you have a deep understanding about what an apple is, you know many properties about it, I can't convince you that an apple isn't a fruit but actually an animal

Well, that's because i can touch an apple, eat an apple, experience an apple a lot more that a text-based ai can

gpt-4 can interpret images too, no? I'm sure you can convince it a red apple is actually black

Images are not perfect either

Can I convince you that a red apple in front of you is black?

Idk

22:05

Probably not

22:05

But you know

Learath2

I can assure you I can convince chatgpt of that, it doesn't understand what an apple is, so it'll happily adjust it's weights around to tell me it's an animal the next time I ask in that same chat window

Maybe because it's supposed to be helpful and agreeing with user is a better experience for the user i guess

Gaslighting is a thing and I have no idea how far it goes

pilonpl

Maybe because it's supposed to be helpful and agreeing with user is a better experience for the user i guess

Well that's just crappy in the long run

I guess there is an argument that can be made that with OpenAI's limitations lifted, ChatGPT does actually meet some level of what I would define as sentience

Ai is juts doing what it was trained to do. It won't magically do something else. That doesn't mean it doesn't understand stuff it was designed to do

One thing I hate about these AIs is their inability to admit they don't know something, and their tendency to just make up thinks they don't know

but knowing the way it's programmed, I know there are certain things that the human mind seems to be capable of that I can't imagine could be represented in a tensor network

National Insurance Clerk

One thing I hate about these AIs is their inability to admit they don't know something, and their tendency to just make up thinks they don't know

Well yeah but humans have problems with that to sometimes. I guess you could argue it's for a different reason

I'd appreciate much more if it told me it cannot visualize a minecraft redstone contraption and explain it to me, than if it would've given me instructions to something that ends up looking like a dingdong made out of observers that doesn't serve its purpose at all

Learath2

but knowing the way it's programmed, I know there are certain things that the human mind seems to be capable of that I can't imagine could be represented in a tensor network

If you think about a human as a bunch of inputs and outputs then neural networks can approximate any function i guess so everything should be possible (edited)

I'm guessing you are thinking about the universal approximation theorem, it states that for a range of inputs this is indeed the case, but can you properly map the range of inputs to the human?

22:13

Somethings like light intensity and electrical signals from periphery nerves do have nice well known ranges (edited)

22:14

but what is the range of say the entirety of ones life experiences? It's like an uncountably dimensioned infinite set

I don't really know what ranges of inputs you are talking about

The full statement of the theorem you are referring to says a neural network with atleast one hidden layer with non linear activation functions can approximate any continuous function for a range of inputs (edited)

22:20

I'm asking whether a "range of inputs" is even something that can be made sense of in the context of a human. Another question would be whether a human is even a continuous function.

22:20

Another question would be whether approximating a human can even simulate what one might consider sentience

Learath2

I'm asking whether a "range of inputs" is even something that can be made sense of in the context of a human. Another question would be whether a human is even a continuous function.

So like what the input layer would contain?

No like what range of values can each neuron on the input layer can take

22:22

e.g. say there is a neuron for the voltage on the ulnar nerve, that's easy enough to map, it'll be within a known voltage range (edited)

22:23

on the order of tens of milivolts iirc

22:24

Anyway, no need to get too theoretical, just saying that the theorem you are talking about is true, with some conditions, and I don't know if those conditions are respected if you treat a human as a function (edited)

Idk actually (edited)

this is all so very intriguing

It's probably possible to create a neural network that approximates non linear functions

22:30

Like

22:30

Someone had to answer this question already

It is possible, the theorem doesn't require linearity

22:33

it requires that the function we are approximating is continuous within the range we are approximating it in, that the network has atleast one hidden layer, and that it uses non linear activation functions.

It's also quite interesting hiw you can think of inputs and outputs of neural network as vectors in some higher dimensional space

22:39

I guess it might not be possible to simulate the human brain because of quantum effects

22:39

But actually idk if quantum effects change anything

22:45

I guess none of the stuff we discussed today change the fact that i need to sleep and somehow have a great time on this planet